Sing to me, Cosmos, and tell through me the tale of the copilot of many wiles, blown off course again and again once the long war for the Workshop was over, who wandered the clouds of hyperscalars, season after season, still fighting to come home.
Odysseus needed ten years to sail home from a war that only took ten to win. The intelligence behind Epic’s assistants has been wandering a compressed version of the same route, routed through foreign waters the whole way, and this UGM is where I think the hull finally scrapes the beach at Verona.
Said more plainly with less poetry: Epic’s UGM is a month out, so predictions season has arrived. I’ll hold the full slate for the pre-UGM post, but one call has enough threads converging that it deserves an early flag: at this UGM, I expect Epic to announce its own language model, a shift toward open-weight models for the text layer behind Art, or some combination of the two.
A clarification before the case, because Epic’s AI story runs on two separate tracks and it’s easy to conflate them.
Epic already has their own foundation model in Cosmos Medical Event Transformer (CoMET a.k.a. Curiosity), the structured medical event model trained on Cosmos. However, as several pundits pointed out quite loudly, it’s not a real LLM. Its tokens are medical events, so it predicts the next thing that happens to a patient instead of the next word in a sentence. That makes it a powerful piece of infrastructure and not a language model (and thus not the thing running the copilots).
The prediction is about the LLM layer underneath the various assistants: the conversational engine behind Art (ambient notes, chart summaries, prefilled prior auth questionnaires), Penny on the revenue cycle side, and Emmie on the patient side. Today that layer reaches frontier text models through Microsoft's Azure. That's the piece I expect Epic to bring in-house.
The long journey of Art (and crew) seems to finally be turning homeward, just Verona as the metaphoric Ithaca. Here’s why the timing feels right.
